Towards improving speech detection robustness for speech recognition in adverse conditions

Recognition performance decreases when recognition systems are used over the telephone network, especially wireless network and noisy environments. It appears that non-efficient speech/non-speech detection (SND) is an important source of this degradation. Therefore, speech detection robustness to noise is a challenging problem to be examined, in order to improve recognition performance for the very noisy communications. Several studies were conducted aiming to improve the robustness of SND used for speech recognition in adverse conditions. The present paper proposes some solutions aiming to improve SND in wireless environment. Speech enhancement prior detection is considered. Then, two versions of SND algorithm, based on statistical criteria, are proposed and compared. Finally, a post-detection technique is introduced in order to reject the wrongly detected noise segments. 2002 Elsevier Science B.V. All rights reserved.